Feature-Driven Road Accident Risk Prediction Using Gradient Boosting Regression

Abstract

Road traffic accidents remain a major global cause of fatalities and economic loss, posing significant challenges to public  safety and urban mobility. Despite advancements in vehicle safety systems and road infrastructure, accurately predicting accident risk remains a complex task that requires advanced analytical techniques. This study develops a predictive framework using Gradient Boosting Regressor (GBR), an ensemble machine learning algorithm, to estimate road accident risk based on environmental and infrastructural features. The analysis incorporates multiple factors, including road type, number of lanes, road curvature, speed limits, lighting conditions, weather patterns, road signage, and historical accident records. The dataset used in this study contains more than 517,000 road condition observations with 13 predictive features and was obtained from the Kaggle Playground Series. During preprocessing, duplicate records were removed to ensure data quality, numerical variables were normalized using Min Max Scaler, and categorical variables were encoded systematically for model compatibility. Experimental results demonstrate that the proposed GBR model achieves strong predictive performance with an accuracy of approximately 88.57% in estimating accident risk levels across diverse road conditions. The findings highlight the significant influence of environmental and infrastructural factors on accident risk and demonstrate the potential of machine learning–based approaches in transportation safety analysis. The proposed framework can assist transportation authorities and policymakers in identifying high-risk road segments and implementing targeted safety interventions to reduce accident occurrences.